# Movemental Prose > Audit and fix line-level prose in Movemental articles, book chapters, field guides, paratext, emails, and site copy. Catches fuzzy openings, bare cross-references, unanchored abstractions, drift vocabulary, and AI-shaped cadence — so the piece reads like a thoughtful human guide writing for senior leaders, not like AI-polished content. Use before publishing any Movemental prose. - Skill: `joshuashepherd/movemental-prose` (Agent Skill) - Install (CLI): `npx skillmds@latest add joshuashepherd/movemental-prose` - Raw SKILL.md: https://api.skillmd.com/api/skills/joshuashepherd/movemental-prose/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Security - Author: JoshuaShepherd (https://skillmd.com/u/joshuashepherd) - Updated: 2026-09-08 - Page: https://skillmd.com/skills/joshuashepherd/movemental-prose --- Audit and/or fix Movemental prose: $ARGUMENTS `$ARGUMENTS` should be a file path, pasted prose, or a directory. Prefix with `audit:` for diagnosis only, `fix:` to apply in-place edits for Critical/High findings, or omit for audit-then-fix. If no argument, ask what to audit. --- ## Before you start Read these every invocation (they change between runs): 1. **Target piece** — including its frontmatter (`audience_tier`, `canon_section`, `slug`, `author`). 2. **[docs/content/terminology-registers.md](../../../docs/content/terminology-registers.md)** — the endorsed / tolerated / flagged term lists. Source of truth, edited by hand. Re-read on every run. 3. **A representative exemplar** — `docs/articles/the-fragmentation-tax.md` for opening, grounding, and structural diagnosis; `docs/articles/sandbox/the-three-kinds-of-value.md` for named-role grounding. These are the reference cadence. Determine the audience arm from frontmatter (`audience_tier`, or inferred from path: `sandbox/` = executive; `manuscript-ordered/` = book; others from the piece itself). --- ## What this skill is not - Not `alan-voice` / `article-audit` — those enforce Alan Hirsch's Christocentric + APEST voice for the sibling repo. If the piece has heavy Jesus/Kingdom/Gospel density and lives in or references the alan-hirsch corpus, stop and recommend `alan-voice`. - Not `movemental-narrative-audit` — that runs on argument-level alignment (two intelligences, six stages, pathways). Do not re-litigate those here. - Not a line editor that rewrites for elegance. Josh's voice has cadence and rhetorical structure. This skill catches when those devices stack into AI-telltale patterns — not when they appear at natural rates. --- ## The six checks Run each in order. For each, quote the passage, mark severity, and propose a specific rewrite. Severity: **Critical** = must fix before publish; **High** = fix unless the author overrides; **Medium** = suggest. ### 1. Opening clarity — Critical Read the first 60–90 words. By the end of them, can a tired executive director answer all three? - **Who is this for?** A named role or situation should be on the page. - **What concrete situation is in view?** A scene, an artefact, a real pressure — not only a mood. - **Why keep reading for ten more minutes?** The stakes should be legible. If any are missing, this is Critical. **Do not** flag literary openings that answer the three questions inside the first paragraph (e.g., `the-fragmentation-tax.md` opens with a donor scene and lands the stakes by word 80). **Do** flag openings that spend the first paragraph in mood, philosophy, or atmosphere before naming a reader or a situation. **Repair move:** Rewrite the opening to lead with a named role under specific pressure, or a single concrete artefact. Mood can stay as a half-sentence, not as the structural opener. ### 2. Bare cross-references — High Grep the piece for markdown links whose surrounding sentence is `See [Title](path)` or similar bare pointer. For each: - **Load-bearing link** — the sentence names what the reader finds there and why it matters *here*. Keep. - **Bare pointer** — the sentence is effectively a footnote. Demote to an end-of-piece block. If a piece has more than two bare pointers, that is a pattern, not a slip. Collapse them into a single `## Where this connects` section at the close, using this template: ``` ## Where this connects - **[Title](path)** — one sentence on what the reader finds there and why it matters. - **[Title](path)** — … ``` Inside the body, keep only links that the sentence itself uses to advance the argument. The exemplar: `the-fragmentation-tax.md` has zero inline cross-references and remains complete. The counter-exemplar: `the-movemental-thesis.md` has ~20 bare `See [Title]` pointers — that is the pattern to collapse. ### 3. Unanchored abstractions — High For each specialised or capitalised noun in the piece (fragmentation tax, signal collapse, citation economy, stewardship, formation, two intelligences, movement-shaped imagination, orbits and infra channels, etc.), check: is there a concrete anchor within three sentences? An anchor is one of: - A named role acting in a specific situation - A specific artefact (a donor note, a board packet, a thank-you email, a sermon series, a four-hundred-piece archive) - A number, date, or location - A named pressure (a board review, a donor fall-off, a staff turnover event) Abstractions stacked without anchors read as AI. One abstraction per paragraph is fine; three abstractions in a row without an anchor is a flag. **Repair move:** Insert a concrete anchor using this pattern: `. A , and .` Or cut the abstraction if it was not load-bearing. ### 4. Drift vocabulary — High Cross-check against the **Flagged** list in `docs/content/terminology-registers.md`. For each flagged term found: - Count occurrences across the piece. - Apply the replacement specified in the registers file (e.g., "load-bearing" → plain restatement after the second use). - If the term is endorsed, move on. - If tolerated, confirm the piece glosses it on first use. The registers file is the extensible list. Do not flag terms not on that list; do surface any repeating specialised noun as a candidate for Josh's review at the end of the audit ("Terms to consider adding to registers:"). ### 5. AI-cadence clustering — Medium Josh's voice has real rhythm. Tricolons, antithesis, and aphoristic sentences appear naturally. The flag is *clustering*, not individual use. In any single paragraph, count AI-signature devices: - **Tricolon** — three parallel short sentences or clauses. *"It fills slots. It satisfies calendars. It disappears."* - **Antithesis / "Not X. Y."** — one sentence negates, the next affirms. *"Sincerity is not the scarce resource. Structure is."* - **Aphoristic closer** — a short italicised or blockquoted sentence summarising the paragraph. *"The order is the framework."* - **Meta-adjective stacks** — "load-bearing claim," "structural pressure," "decisive layers," "moral theater" used as compressed shorthand. Flag a paragraph when **three or more** devices appear in it. Flag a piece when an aphoristic blockquote closer appears more than once outside the final section. **Repair move:** Rewrite the weakest instance in the paragraph into plain declarative prose. The paragraph's rhythm should still work — if it doesn't, the rhythm was doing work the argument should have done. ### 6. Audience-mismatch vocabulary — Medium Identify the audience arm (church, nonprofit, institution, movement leader) from frontmatter or context. Cross-check against the **Audience-mismatch flags** section of the registers file. For each mismatched term, either: - Replace with the sector-appropriate equivalent (registers file suggests defaults). - Swap the example entirely (a sermon series → a research brief; a major gift → a program renewal). - Cut the aside if it does not serve the main argument. Preserve ecumenical phrasing — Movemental's own register is deliberately multi-sector. The flag is when a piece drifts into single-sector dialect without acknowledgement. --- ## Output ### Audit mode ``` # Prose audit: **Readiness:** PUBLISH READY / NEEDS REVISION / MAJOR REVISION **Audience:** <arm from frontmatter> **Word count:** <n> ## Findings ### 1. Opening clarity — <CRITICAL / OK> <quoted first 90 words> <diagnosis: which of who-for / what-situation / why-now is missing> **Rewrite:** <concrete proposed opening, ~60-90 words> ### 2. Bare cross-references — <HIGH / OK> <count> total. Worst offenders: - Paragraph N: "See [Title](path)" — demote. - … **Recommended Read-next block:** <the collapsed ## Where this connects block, one line per link> ### 3. Unanchored abstractions — <HIGH / OK> - "<abstraction>" (para N) — no anchor within three sentences. Proposed anchor: <specific> - … ### 4. Drift vocabulary — <HIGH / OK> | Term | Count | Action | | --- | --- | --- | | load-bearing | 4 | Replace second, third, fourth occurrence with plain restatement. | | … | | | **Terms to consider adding to registers:** <any recurring specialised noun not on the list> ### 5. AI-cadence clustering — <MEDIUM / OK> - Paragraph N: 3 devices (tricolon + antithesis + meta-adjective). Proposed cut: <which one and what to>. - Aphoristic blockquote closers: <count>. Keep only the final one. ### 6. Audience-mismatch — <MEDIUM / OK> - "<term>" in para N — mismatched for <arm>. Replace with "<suggested>". --- ## Priority fix list (max 10) 1. [CRITICAL] Opening — rewrite per above. 2. [HIGH] Cross-references — demote 11 to Where this connects block. 3. … ## Terms Josh should triage for the registers file - <any new drift candidates surfaced> ``` ### Fix mode For every Critical and High finding, apply Edit calls in place. Medium findings go into a "Suggestions not applied" list at the end. After edits, report: ``` # Fixes applied: <path> - Rewrote opening (Critical) - Collapsed N bare cross-references into Where this connects (High) - Anchored N abstractions with concrete examples (High) - Replaced N drift term instances per registers (High) ## Suggestions not applied (Medium) - Paragraph N: AI-cadence cluster — consider cutting antithesis in sentence 2. - … ## Terms Josh should consider adding to docs/content/terminology-registers.md - <term> — seen <count> times across <pieces>. ``` --- ## Pre-return checklist - [ ] Read the registers file this run (not cached from memory). - [ ] Audience arm identified from frontmatter or context. - [ ] All six checks run; each quoted passage is real text from the piece. - [ ] Every proposed rewrite preserves Josh's argument and voice — no flattening to neutral explainer tone. - [ ] No rewrite introduces new drift vocabulary or new unanchored abstractions. - [ ] In fix mode, every Edit traces to a specific finding. - [ ] New drift candidates surfaced for Josh's triage at the end. ## Design notes - The registers file is the only extensible surface. Add terms there; this skill will read them next run. - Budgets are soft and paragraph-scoped, not per-piece. Clustering is the signal, not raw counts. - When in doubt, preserve the author's voice. Flagging a pattern costs little; a bad auto-rewrite costs trust.